What is AI fine-tuning?
Fine-tuning means further training an existing model on examples for a particular task or behavior. It changes the model’s learned parameters. The aim might be a more consistent format, style, or task response; improvement must be measured rather than assumed.
Fine-tuning, prompting, and retrieval
Prompting supplies instructions at the time of a request. Retrieval supplies selected source material for the model to use. Fine-tuning changes the model through additional training. These approaches address different problems and can be combined.
When is fine-tuning worth considering?
Start with a clear task and examples of successful and unsuccessful results. Test a well-written prompt first. If performance remains inconsistent across repeated tasks, a training approach may be worth investigating. For changing facts or a collection of documents, consider whether supplying the relevant sources better addresses the need.
How do you evaluate the result?
Keep evaluation examples separate from training examples. Compare the original and adapted model on accuracy, consistency, and the errors that matter in the workflow. Better results on one task do not establish reliability for every task.
